- Email: [email protected]
Examining the Association Between School Vending Machines and Children's Body Mass Index by Socioeconomic Status
Research Brief Examining the Association Between School Vending Machines and Children’s Body Mass Index by Socioeconomic Status Jeffrey K. O'Hara, PhD; Lindsey Haynes-Maslow, PhD, MHA ABSTRACT Objective: To examine the association between vending machine availability in schools and body mass index (BMI) among subgroups of children based on gender, race/ethnicity, and socioeconomic status classifications. Methods: First-difference multivariate regressions were estimated using longitudinal fifth- and eighthgrade data from the Early Childhood Longitudinal Study. The specifications were disaggregated by gender, race/ethnicity, and family socioeconomic status classifications. Results: Vending machine availability had a positive association (P < .10) with BMI among Hispanic male children and low-income Hispanic children. Living in an urban location (P < .05) and hours watching television (P < .05) were also positively associated with BMI for these subgroups. Supplemental Nutrition Assistance Program enrollment was negatively associated with BMI for low-income Hispanic students (P < .05). These findings were not statistically significant when using Bonferroni adjusted critical values. Conclusions and Implications: The results suggest that the school food environment could reinforce health disparities that exist for Hispanic male children and low-income Hispanic children. Key Words: childhood obesity, schools, minority health (J Nutr Educ Behav. 2015;47:526-531.) Accepted August 3, 2015. Published online September 12, 2015.
INTRODUCTION Obesity rates are highest among children from racial/ethnic minority1 and lower socioeconomic backgrounds.2,3 Understanding whether the school food environment, which includes school meal programs and competitive foods sold outside meal programs, reinforces existing weight disparities is a priority because individual, interpersonal, and environmental influences on children's dietary choices vary by gender, race/ethnicity, and family socioeconomic status.4 Students from racial/ethnic minority groups and low socioeconomic backgrounds are more likely to consume sweetened beverages or purchase snacks and bever-
ages from school vending machines instead of school lunch.5-7 It is important to estimate whether there is an association between vending machine availability in schools and children's weight because historically, the most frequently purchased foods in school vending machines had minimal nutritional value, such as soft drinks, chips, and candy.8-10 Between 2005 and 2006, the US Health Behavior in School Aged Children survey found that soft drinks were the most common item used to stock vending machines.8 This raises public health concerns because they have been linked with weight gain,11 metabolic disorders,12 and dental caries.13
Food and Environment Program, Union of Concerned Scientists, Washington, DC Conflict of Interest Disclosure: The authors’ conflict of interest disclosures can be found online with this article on www.jneb.org. Address for correspondence: Lindsey Haynes-Maslow, PhD, MHA, Union of Concerned Scientists, 1825 K St NW, Ste 800, Washington, DC 20006; Phone: (202) 331-5432; Fax: (202) 223-6162; E-mail: [email protected] Ó2015 Society for Nutrition Education and Behavior. Published by Elsevier, Inc. All rights reserved. http://dx.doi.org/10.1016/j.jneb.2015.08.001
526
Studies examining associations between school vending machines and weight gain have mixed findings.14-17 The study by Van Hook and Altman17 is the only one among these that reported results by children's gender, race/ethnicity, and socioeconomic status, and it found no association between exposure to less nutrient-dense food and children's body mass index (BMI) for these subgroups.17 However, Van Hook and Altman did not report testing for significance among disaggregated subsets of these populations, such as low-income females or Hispanic males. The purpose of this study was to estimate the association of school vending machine availability and BMI among subgroups of children based on gender, race/ethnicity, and socioeconomic status at a greater level of specificity than previously examined in the literature.
METHODS Dataset The authors used the Early Childhood Longitudinal Study, Kindergarten Class of 1998–99 (ECLS-K) panel dataset,
Journal of Nutrition Education and Behavior Volume 47, Number 6, 2015
Journal of Nutrition Education and Behavior Volume 47, Number 6, 2015 which has been used in previous studies.16,17 The ECLS-K was administered by the National Center for Education Statistics at the US Department of Education and followed a nationally representative cohort of children from kindergarten through eighth grade, tracking the same children over time as they aged through the school system. Data on children's demographics, diet, physical activity, and weight were collected from children, parents, teachers, and school administrators. Children's height and weight were directly measured by study staff in all survey rounds. The authors used data from the fifth- and eighth-grade surveys, because these were years when school administrators were asked about vending machine availability. These surveys were administered in 2004 and 2007, respectively. Institutional review board approval was not required for this study because the version of the ECLS-K dataset used by the authors is publicly available and does not reveal confidential information that can be identified to a particular child.
Measures The outcome variable was the change in children's BMI between fifth and eighth grades. The explanatory variable of interest was a variable indicating a change in whether students could purchase food or beverages at vending machines at school between fifth and eighth grades. The authors controlled for child-, household-, and school-level characteristics. Child-level characteristics included changes in the number of times per week the child ate breakfast with the family, ate dinner with the family, and exercised for 20 consecutive minutes; the number of hours per week the child watched television; and whether the child had a disability. Because BMI changes between the fifth and eighth grades could be attributed to changes in physiology from aging or other timevarying unobserved external influences, the authors also included a constant. Household-level characteristics included changes in the number of individuals living in the household; whether the child lived with nonbio-
logical parents, a single mother, and a single father; whether the family received Supplemental Nutrition Assistance Program benefits; and the family's ratio of income to federal poverty guidelines (FPG). The authors used midpoints from the categorical variable of annual household income to represent the family's income, with families in the highest income category ($ $200,001) categorized as having an income of $200,001. The authors used household income and household size to calculate the ratio of family income to FPG for 2004 and 2007.18,19 School-level characteristics included indicator variables reflecting a change in whether the school was located in urban (large/midsize city) and rural (small town) regions, with children living in a suburban region as the reference variable, as well as the percentage of minority students enrolled at school.
Data Analysis To estimate whether the change in vending machine availability was associated with a change in BMI, the authors used first difference equations. These represent the lagged value of the observations (t – 1) subtracted from their current period value (t). In this equation, BMI is represented by y and the independent variables are represented by x. For each period t, the vector y has dimension nx1, where n is the number of students and the matrix x has dimension nxk, where k represents the number of independent variables. The vector of coefficients, b, has dimension kx1 and was estimated using ordinary least squares. First difference equations were used to control for the possibility that unobserved time-invariant individual effects may be correlated with the error term, ε:
yt yt1 ¼ ðxt xt1 Þb þ εt εt1 The authors estimated this equation for all male students, all female students, and all low-income students, in addition to 6 subgroups each for white, African American, and Hispanic children: all children, males, females, all children from low-income families, males from low-income families, and
O’Hara and Haynes-Maslow 527 females from low-income families. This totals 21 specifications. Children were classified as being from lowincome families if their family income was below 185% of FPG in both fifth and eighth grades. Vending machines may be more likely to exist in schools in which there is a greater demand for them, and children's BMI could be systematically higher in such schools for reasons unrelated to vending machines. If so, vending machine availability would be an endogenous variable. The authors tested whether the change in vending machine availability was an endogenous variable following a similar procedure used in previous research on this topic.16,17 The researchers used robust standards errors to control for heteroscedasticity. Results were reported at the 5%, 10%, and 0.2% statistical significance levels. The latter critical value represents a Bonferroni correction equal to ratio of the .05 significance level and the number of specifications (0.002 ¼ .05/21). The significance of parameter estimates with Bonferroni corrected critical values is reported because the probability of a Type I error (false positive) increases as a greater number of specifications are estimated from the same dataset. Subjects' data were eliminated that had missing values for included variables in either the fifth or eighth grade from the sample. Data analysis was conducted in Stata (version 13, StataCorp, College Station, TX, 2013).
RESULTS The final sample included 2,263 students (Table 1). A total of 48% of students were female, 64% were white, 11% were African American, and 17% were Hispanic. The average BMI of students in the sample increased from 20.6 to 23.0 between the fifth and eighth grades. The percentage of overweight and obese children increased during this period from 26% and 12%, respectively, to 39% and 22%, respectively. Twentyseven percent of fifth-graders had access to school vending machines, compared with 63% of eighth-graders. Table 2 shows that the ability of students to purchase sweets, salty snacks, and sweetened beverages (ie,
Journal of Nutrition Education and Behavior Volume 47, Number 6, 2015
528 O’Hara and Haynes-Maslow
Table 1. Descriptive Statistics for Fifth- and Eighth-Grade Students (n ¼ 2,263)
Characteristics Child-level characteristics Child body mass index (mean SD) Overweight Obese Female White African American Hispanic Has disability Hours of television watching/wk (mean SD) Days/wk exercises 20 min (mean SD) Household characteristics Lives with nonbiological parents Lives with single mother Lives with single father Individuals living in household, n (mean SD) Family receives Supplemental Nutrition Assistance Ratio of income/poverty line (mean SD) Days/wk eat breakfast with family (mean SD) Days/wk eat dinner with family (mean SD) School characteristics Has vending machines Urban Rural Percent minority students in school (mean SD)
soda/sports drinks/fruit drinks) at school increased between fifth and eighth grades from 50% to 69%, 46% to 74%, and 39% to 64%, respectively. Furthermore, the percentage of students who purchased sweets, salty snacks, and sweetened beverages in school and usually made such purchases from school vending machines increased between fifth and eighth grade from 5% to 13%, 9% to 14%, and 47% to 53%, respectively. Thus, there were percent increases of 277%, 197%, and 129% of students in the sample who predominately purchased sweets, salty snacks, and sweetened beverages, respectively, at school vending machines between fifth and eighth grades. For brevity, Table 3 presents results from 4 select multivariate regressions: all Hispanic students (model 1), Hispanic male students (model 2), all low-income Hispanic students (model 3), and low-income Hispanic male students (model 4). The latter 3 specifica-
Fifth Grade (n ¼ 2,263) n (%)
Eighth Grade (n ¼ 2,263) n (%)
20.6 4.4 598 (26.4) 269 (11.9) 1,077 (47.6) 1,443 (63.8) 249 (11.0) 391 (17.3) 393 (17.4) 7.4 4.0 3.8 1.9
23.0 4.8 888 (39.3) 497 (22.0) 1,077 (47.6) 1,443 (63.8) 249 (11.0) 391 (17.3) 384 (17.0) 7.5 5.9 5.4 1.8
331 (14.6) 451 (19.9) 54 (2.4) 4.5 1.3 312 (13.8) 2.7 2.2 3.4 2.4 5.5 1.7
388 (17.1) 462 (20.4) 52 (2.3) 4.4 1.3 307 (13.6) 2.7 2.1 3.0 2.2 5.2 1.7
605 (26.7) 625 (27.6) 809 (35.7) 32.6 27.8
1,433 (63.3) 604 (26.7) 788 (34.8) 33.5 26.8
tions are presented because these correspond to the specifications in which vending machine access was statistically significant above a 90% confidence level, and model 1 is reported as a benchmark. Vending machine availability was not endogenous in each of the reported specifications. The F tests for joint significance of all of the variables in the 4 regressions were statistically significant (P < .05). The constant was the only statistically significant (P < .002) parameter estimate in the 4 reported specifications when using Bonferroni corrected critical values. Other variables were statistically significant when using less conservative critical values. An increase in vending machine availability was associated with an increase in BMI by 0.46 units for Hispanic male children (P < .10), 0.40 for low-income Hispanic children (P < .10), and 0.76 for low-income Hispanic male chil-
dren (P < .05). Living in an urban location was significant and positive in models 1 (P < .05), 2 (P < .05), and 3 (P < .05). Supplemental Nutrition Assistance Program enrollment had a negative association with BMI in models 1 (P < .05) and 3 (P < .05). Hours watching television had a positive association with BMI in models 2 (P < .05), 3 (P < .05), and 4 (P < .05). In model 2, exercise had a positive association (P < .10) and breakfast a negative association (P < .10) with BMI. The other variables were not statistically significant in any of the 4 reported specifications.
DISCUSSION The association between vending machine availability and BMI was statistically insignificant when students were grouped at aggregated levels by gender, race/ethnicity, or socioeconomic status. This is consistent with findings
Journal of Nutrition Education and Behavior Volume 47, Number 6, 2015
O’Hara and Haynes-Maslow 529
Table 2. Sweets, Salty Snacks, and Sweetened Beverage Availability and Purchases in Fifth and Eighth Grades
Food Type and Purchase
Fifth Grade n (%)
Eighth Grade n (%)
Sweets Can purchase sweets at school Times bought sweets at school, n/wk (mean SD)a Purchased sweets in school vending machinesb
1,138 (50.3) 1.65 3.55 26 (4.6)
1,533 (68.5) 1.46 3.16 98 (12.5)
Salty snacks Can purchase salty snacks at school Times bought salty snacks at school, n/wk (mean SD)a Purchased salty snacks in school vending machinesb
1,150 (46.4) 1.01 2.76 31 (8.6)
1,635 (73.8) 1.20 2.98 92 (14.0)
Sweetened beverages Can purchase soft drinks, fruit drinks, and sports drinks at school Times bought soft drinks, fruit drinks, and sports drinks at school, n/wk (mean SD)a Purchased soft drinks, fruit drinks, and sports drinks in school vending machinesb
882 (39.0) 0.94 2.58 131 (46.6)
1,423 (64.3) 1.09 2.57 300 (52.8)
a
Number of times food purchased per week at school is calculated for children who could purchase those foods at school; Purchased foods from school vending machines are calculated for students who purchased those foods at school.
b
by Van Hook and Altman.17 The predicted association between vending machine availability and higher BMI levels had greater statistical significance among Hispanic male and low-income children than other socioeconomic subgroups, although the finding among Hispanic subgroups
meals, likely owing to a shortage of time for purchasing and preparing food,20 language barriers that impede both children's and parents' dietary education,21 and taste preferences for nutrient-deficient foods.22 Hispanic male children may be more vulnerable to increases in BMI than are Hispanic
was insignificant with Bonferroni corrected critical values. Hispanic children from lowerincome families may have a greater association between vending machine availability and BMI than other subgroups because of challenges their parents confront preparing healthier
Table 3. First Difference Regressions for Change in Student’s BMI From Fifth to Eighth Grade Model 1 Hispanic Students (n ¼ 390) b Coefficient (SD) 2.46 (0.14)***
Model 2 Hispanic Male Students (n ¼ 210) b Coefficient (SD) 2.28 (0.20)***
Vending machine availability (categorical)
0.24 (0.20)
0.46 (0.24)*
0.40 (0.23)*
0.76 (0.33)**
Television watching, h/wk
0.03 (0.30)
0.05 (0.03)**
0.05 (0.02)**
0.09 (0.03)**
Independent Variables Constant
Exercises 20 min/d, d/wk Family receives SNAP (categorical) Eat breakfast with family, d/wk Urban (categorical) 2
0.05 (0.04) 0.78** (0.30) 0.02 (0.05) 1.25 (0.42)**
0.10 (0.06)* 0.48 (0.42) 0.10 (0.06)* 1.53 (0.54)**
Model 3 Low-Income Hispanic Students (n ¼ 264) b Coefficient (SD) 2.53 (0.17)***
Model 4 Low-Income Hispanic Male Students (n ¼ 134) b Coefficient (SD) 2.47 (0.30)***
0.04 (0.06)
0.10 (0.09)
0.70** (0.33)
0.49 (0.47)
0.03 (0.06)
0.10 (0.08)
1.35 (0.53)**
1.02 (0.88)
R
0.07
0.13
0.08
0.17
F test P
.01**
.03**
.01**
.02**
*P < .10; **P < .05; ***P < .002. Note: First difference estimates are derived from regression models that include changes in whether the child has a disability, lives with nonbiological parents, lives with a single mother, lives with a single father, days per week that the child eats dinner with the family, the ratio of income to federal poverty line, lives in a rural region, and the percent minority students enrolled in the child’s school as control variables. Parameter estimates for statistically insignificant variables are not reported.
530 O’Hara and Haynes-Maslow female children because, as reported in the ECLS-K dataset, they purchased salty snacks and sweetened beverages with a greater frequency at school. Other findings from this study reveal that television watching may be positively associated with BMI, which is consistent with evidence that sedentary lifestyles can contribute to increases in BMI.23 The positive parameter estimate associated with exercise in males could result from increased muscle mass. Consistent with other studies, living in an urban area may be associated with increased BMI, possibly owing to a lack of recreational venues and safe places to be active.24 Supplemental Nutrition Assistance Program enrollment and eating breakfast with the family had a negative association with BMI, possibly because these factors indicate greater food security and more stable eating patterns. The magnitude of the parameter estimates for the constant was similar across models 1–4 and ranged from 2.3 to 2.5.
IMPLICATIONS FOR RESEARCH AND PRACTICE Because of concern regarding the health impacts of children's diets, federal nutrition standards for competitive foods in schools, including those sold in vending machines, were updated on July 1, 2014. Implemented in the 2014–2015 academic year, nutrition standards for sweetened beverages—the most commonly purchased item in school vending machines—must not contain more than 60 calories per 12-oz serving. This is approximately 40% to 43% of the calories in a nondiet 12-oz soft drink. The health impacts of the new competitive food standards have yet to be evaluated. Although the calorie amounts for beverages are lower, such changes could result in increased sweetened beverage consumption if students' perceptions of serving sizes are altered (eg, fewer calories in a serving size may decrease the guilt of overconsuming calories from sweetened beverages).25 Thus, understanding whether changes in the school food environment have differential associations with chil-
Journal of Nutrition Education and Behavior Volume 47, Number 6, 2015 dren's health status by subgroup remains pertinent and highlights the value of research on this topic estimating associations at specific levels of subgroups.
10.
ACKNOWLEDGMENTS The work was undertaken when both authors were employed at the Union of Concerned Scientists.
REFERENCES 1. Ogden CL, Carroll MD, Kit BK, Flegal KM. Prevalence of childhood and adult obesity in the United States, 2011-2012. JAMA. 2014;311: 806-814. 2. Singh GK, Siahpush M, Kogan MD. Rising social inequalities of US childhood obesity, 2003-2007. Ann Epidemiol. 2010;20:40-52. 3. Skelton JA, Cook SR, Auinger P, Klein JD, Barlow SE. Prevalence and trends of severe obesity among US children and adolescents. Acad Pediatr. 2009; 9:322-329. 4. Sallis JF, Owen N, Fisher EB. Ecological models of heath behavior. In: Glanz K, Rimer BK, Viswanath K, eds. Health Behavior and Health Education: Theory, Research and Practice. 4th ed. San Francisco, CA: Josey-Bass; 2008. 5. Fernandes MM. The effect of soft drink availability in elementary schools on consumption. J Am Diet Assoc. 2008; 108:1445-1452. 6. Briefel RR, Crepinsek MK, Cabili C, Wilson A, Gleason PM. School food environments and practices affect dietary behaviors of U.S. public school children. J Am Diet Assoc. 2009;109(2 suppl):S91-S107. 7. Park S, Sappenfield WM, Huang Y, Sherry B, Bensyl DM. The impact of the availability of school vending machines on eating behavior during lunch: the youth physical activity and nutrition survey. J Am Diet Assoc. 2010; 110:1532-1536. 8. Rover AJ, Nansel TR, Wang J, Lannotti RJ. Foods sold in school vending machines are associated with overall student dietary intake. J Adolesc Health. 2011;48:13-19. 9. O’Toole TP, Anderson S, Miller C, Guthrie J. Nutrition services and foods and beverages available at school: re-
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
sults from the School Health Policies and Programs Study 2006. J Sch Health. 2007;77:500-521. Fox MK, Gordon A, Nogales R, Wilson A. Availability and consumption of competitive foods in US public schools. J Am Diet Assoc. 2009; 109(suppl 2):S57-S66. US Department of Agriculture and US Dept of Health and Human Services. Dietary Guidelines for Americans, 2010. 7th ed. Washington, DC: US Government Printing Office; 2010. Hu FB, Malik VS. Sugar-sweetened beverages and risk of obesity and type 2 diabetes: epidemiological evidence. Physiol Behav. 2010;100:47-54. Marshall TA, Levy SM, Broffitt B, et al. Dental caries and beverage consumption in young children. Pediatrics. 2003;112:e184-e191. Larson N, Story M. Are ‘‘competitive foods’’ sold at school making our children fat? Health Affairs. 2010;29: 430-435. Anderson PM, Butcher KM. Reading, writing, and refreshments: are school finances contributing to children’s obesity? J Human Resources. 2006;41: 467-494. Datar A, Nicosia N. Junk food in schools and childhood obesity. J Policy Analysis Management. 2012;31:312-337. Van Hook J, Altman CE. Competitive food sales in schools and childhood obesity: a longitudinal study. Sociol Educ. 2012;85:23-29. US Department of Health and Human Services. The 2007 HHS poverty guidelines. http://aspe.hhs.gov/poverty/07pov erty.shtml. Accessed November 15, 2014. US Department of Health and Human Services. The 2004 HHS poverty guidelines. http://aspe.hhs.gov/poverty/04po verty.shtml. Accessed November 3, 2014. Haynes-Maslow L, Parsons SE, Wheeler SB, Leone LA. Understanding barriers to fruit and vegetable consumption among low-income individuals: a qualitative study. Prev Chronic Dis. 2013;10:1202-1206. Gray VB, Cossman JS, Dodson WL, Byrd SH. Dietary acculturation of Hispanic immigrants in Mississippi. Salud Publica Mex. 2005;47:351-360. Batis C, Hernandez-Barrera L, Barquera S, Rivera JA, Popkin BM. Food acculturation drives dietary differences among Mexicans, Mexican
Journal of Nutrition Education and Behavior Volume 47, Number 6, 2015 Americans, and non-Hispanic whites. J Nutr. 2011;141:1898-1906. 23. Boone JE, Gordon-Larsen P, Adair LS, Popkin BM. Screen time and physical activity during adolescence: longitudinal
effects on obesity in young adulthood. Int J Behav Nutr Phys Act. 2007;4:26. 24. Wolch J, Jerrett M, Reynolds R, et al. Childhood obesity and proximity to urban parks and recreational resources:
O’Hara and Haynes-Maslow 531 a longitudinal cohort study. Health Place. 2011;17:207-214. 25. Wansink B, Chandon P. Can ‘‘low-fat’’ nutrition labels lead to obesity? J Market Res. 2006;43:605-607.
531.e1 O’Hara and Haynes-Maslow
CONFLICT OF INTEREST The authors have not stated any conflicts of interest.
Journal of Nutrition Education and Behavior Volume 47, Number 6, 2015
INTRODUCTION Obesity rates are highest among children from racial/ethnic minority1 and lower socioeconomic backgrounds.2,3 Understanding whether the school food environment, which includes school meal programs and competitive foods sold outside meal programs, reinforces existing weight disparities is a priority because individual, interpersonal, and environmental influences on children's dietary choices vary by gender, race/ethnicity, and family socioeconomic status.4 Students from racial/ethnic minority groups and low socioeconomic backgrounds are more likely to consume sweetened beverages or purchase snacks and bever-
ages from school vending machines instead of school lunch.5-7 It is important to estimate whether there is an association between vending machine availability in schools and children's weight because historically, the most frequently purchased foods in school vending machines had minimal nutritional value, such as soft drinks, chips, and candy.8-10 Between 2005 and 2006, the US Health Behavior in School Aged Children survey found that soft drinks were the most common item used to stock vending machines.8 This raises public health concerns because they have been linked with weight gain,11 metabolic disorders,12 and dental caries.13
Food and Environment Program, Union of Concerned Scientists, Washington, DC Conflict of Interest Disclosure: The authors’ conflict of interest disclosures can be found online with this article on www.jneb.org. Address for correspondence: Lindsey Haynes-Maslow, PhD, MHA, Union of Concerned Scientists, 1825 K St NW, Ste 800, Washington, DC 20006; Phone: (202) 331-5432; Fax: (202) 223-6162; E-mail: [email protected] Ó2015 Society for Nutrition Education and Behavior. Published by Elsevier, Inc. All rights reserved. http://dx.doi.org/10.1016/j.jneb.2015.08.001
526
Studies examining associations between school vending machines and weight gain have mixed findings.14-17 The study by Van Hook and Altman17 is the only one among these that reported results by children's gender, race/ethnicity, and socioeconomic status, and it found no association between exposure to less nutrient-dense food and children's body mass index (BMI) for these subgroups.17 However, Van Hook and Altman did not report testing for significance among disaggregated subsets of these populations, such as low-income females or Hispanic males. The purpose of this study was to estimate the association of school vending machine availability and BMI among subgroups of children based on gender, race/ethnicity, and socioeconomic status at a greater level of specificity than previously examined in the literature.
METHODS Dataset The authors used the Early Childhood Longitudinal Study, Kindergarten Class of 1998–99 (ECLS-K) panel dataset,
Journal of Nutrition Education and Behavior Volume 47, Number 6, 2015
Journal of Nutrition Education and Behavior Volume 47, Number 6, 2015 which has been used in previous studies.16,17 The ECLS-K was administered by the National Center for Education Statistics at the US Department of Education and followed a nationally representative cohort of children from kindergarten through eighth grade, tracking the same children over time as they aged through the school system. Data on children's demographics, diet, physical activity, and weight were collected from children, parents, teachers, and school administrators. Children's height and weight were directly measured by study staff in all survey rounds. The authors used data from the fifth- and eighth-grade surveys, because these were years when school administrators were asked about vending machine availability. These surveys were administered in 2004 and 2007, respectively. Institutional review board approval was not required for this study because the version of the ECLS-K dataset used by the authors is publicly available and does not reveal confidential information that can be identified to a particular child.
Measures The outcome variable was the change in children's BMI between fifth and eighth grades. The explanatory variable of interest was a variable indicating a change in whether students could purchase food or beverages at vending machines at school between fifth and eighth grades. The authors controlled for child-, household-, and school-level characteristics. Child-level characteristics included changes in the number of times per week the child ate breakfast with the family, ate dinner with the family, and exercised for 20 consecutive minutes; the number of hours per week the child watched television; and whether the child had a disability. Because BMI changes between the fifth and eighth grades could be attributed to changes in physiology from aging or other timevarying unobserved external influences, the authors also included a constant. Household-level characteristics included changes in the number of individuals living in the household; whether the child lived with nonbio-
logical parents, a single mother, and a single father; whether the family received Supplemental Nutrition Assistance Program benefits; and the family's ratio of income to federal poverty guidelines (FPG). The authors used midpoints from the categorical variable of annual household income to represent the family's income, with families in the highest income category ($ $200,001) categorized as having an income of $200,001. The authors used household income and household size to calculate the ratio of family income to FPG for 2004 and 2007.18,19 School-level characteristics included indicator variables reflecting a change in whether the school was located in urban (large/midsize city) and rural (small town) regions, with children living in a suburban region as the reference variable, as well as the percentage of minority students enrolled at school.
Data Analysis To estimate whether the change in vending machine availability was associated with a change in BMI, the authors used first difference equations. These represent the lagged value of the observations (t – 1) subtracted from their current period value (t). In this equation, BMI is represented by y and the independent variables are represented by x. For each period t, the vector y has dimension nx1, where n is the number of students and the matrix x has dimension nxk, where k represents the number of independent variables. The vector of coefficients, b, has dimension kx1 and was estimated using ordinary least squares. First difference equations were used to control for the possibility that unobserved time-invariant individual effects may be correlated with the error term, ε:
yt yt1 ¼ ðxt xt1 Þb þ εt εt1 The authors estimated this equation for all male students, all female students, and all low-income students, in addition to 6 subgroups each for white, African American, and Hispanic children: all children, males, females, all children from low-income families, males from low-income families, and
O’Hara and Haynes-Maslow 527 females from low-income families. This totals 21 specifications. Children were classified as being from lowincome families if their family income was below 185% of FPG in both fifth and eighth grades. Vending machines may be more likely to exist in schools in which there is a greater demand for them, and children's BMI could be systematically higher in such schools for reasons unrelated to vending machines. If so, vending machine availability would be an endogenous variable. The authors tested whether the change in vending machine availability was an endogenous variable following a similar procedure used in previous research on this topic.16,17 The researchers used robust standards errors to control for heteroscedasticity. Results were reported at the 5%, 10%, and 0.2% statistical significance levels. The latter critical value represents a Bonferroni correction equal to ratio of the .05 significance level and the number of specifications (0.002 ¼ .05/21). The significance of parameter estimates with Bonferroni corrected critical values is reported because the probability of a Type I error (false positive) increases as a greater number of specifications are estimated from the same dataset. Subjects' data were eliminated that had missing values for included variables in either the fifth or eighth grade from the sample. Data analysis was conducted in Stata (version 13, StataCorp, College Station, TX, 2013).
RESULTS The final sample included 2,263 students (Table 1). A total of 48% of students were female, 64% were white, 11% were African American, and 17% were Hispanic. The average BMI of students in the sample increased from 20.6 to 23.0 between the fifth and eighth grades. The percentage of overweight and obese children increased during this period from 26% and 12%, respectively, to 39% and 22%, respectively. Twentyseven percent of fifth-graders had access to school vending machines, compared with 63% of eighth-graders. Table 2 shows that the ability of students to purchase sweets, salty snacks, and sweetened beverages (ie,
Journal of Nutrition Education and Behavior Volume 47, Number 6, 2015
528 O’Hara and Haynes-Maslow
Table 1. Descriptive Statistics for Fifth- and Eighth-Grade Students (n ¼ 2,263)
Characteristics Child-level characteristics Child body mass index (mean SD) Overweight Obese Female White African American Hispanic Has disability Hours of television watching/wk (mean SD) Days/wk exercises 20 min (mean SD) Household characteristics Lives with nonbiological parents Lives with single mother Lives with single father Individuals living in household, n (mean SD) Family receives Supplemental Nutrition Assistance Ratio of income/poverty line (mean SD) Days/wk eat breakfast with family (mean SD) Days/wk eat dinner with family (mean SD) School characteristics Has vending machines Urban Rural Percent minority students in school (mean SD)
soda/sports drinks/fruit drinks) at school increased between fifth and eighth grades from 50% to 69%, 46% to 74%, and 39% to 64%, respectively. Furthermore, the percentage of students who purchased sweets, salty snacks, and sweetened beverages in school and usually made such purchases from school vending machines increased between fifth and eighth grade from 5% to 13%, 9% to 14%, and 47% to 53%, respectively. Thus, there were percent increases of 277%, 197%, and 129% of students in the sample who predominately purchased sweets, salty snacks, and sweetened beverages, respectively, at school vending machines between fifth and eighth grades. For brevity, Table 3 presents results from 4 select multivariate regressions: all Hispanic students (model 1), Hispanic male students (model 2), all low-income Hispanic students (model 3), and low-income Hispanic male students (model 4). The latter 3 specifica-
Fifth Grade (n ¼ 2,263) n (%)
Eighth Grade (n ¼ 2,263) n (%)
20.6 4.4 598 (26.4) 269 (11.9) 1,077 (47.6) 1,443 (63.8) 249 (11.0) 391 (17.3) 393 (17.4) 7.4 4.0 3.8 1.9
23.0 4.8 888 (39.3) 497 (22.0) 1,077 (47.6) 1,443 (63.8) 249 (11.0) 391 (17.3) 384 (17.0) 7.5 5.9 5.4 1.8
331 (14.6) 451 (19.9) 54 (2.4) 4.5 1.3 312 (13.8) 2.7 2.2 3.4 2.4 5.5 1.7
388 (17.1) 462 (20.4) 52 (2.3) 4.4 1.3 307 (13.6) 2.7 2.1 3.0 2.2 5.2 1.7
605 (26.7) 625 (27.6) 809 (35.7) 32.6 27.8
1,433 (63.3) 604 (26.7) 788 (34.8) 33.5 26.8
tions are presented because these correspond to the specifications in which vending machine access was statistically significant above a 90% confidence level, and model 1 is reported as a benchmark. Vending machine availability was not endogenous in each of the reported specifications. The F tests for joint significance of all of the variables in the 4 regressions were statistically significant (P < .05). The constant was the only statistically significant (P < .002) parameter estimate in the 4 reported specifications when using Bonferroni corrected critical values. Other variables were statistically significant when using less conservative critical values. An increase in vending machine availability was associated with an increase in BMI by 0.46 units for Hispanic male children (P < .10), 0.40 for low-income Hispanic children (P < .10), and 0.76 for low-income Hispanic male chil-
dren (P < .05). Living in an urban location was significant and positive in models 1 (P < .05), 2 (P < .05), and 3 (P < .05). Supplemental Nutrition Assistance Program enrollment had a negative association with BMI in models 1 (P < .05) and 3 (P < .05). Hours watching television had a positive association with BMI in models 2 (P < .05), 3 (P < .05), and 4 (P < .05). In model 2, exercise had a positive association (P < .10) and breakfast a negative association (P < .10) with BMI. The other variables were not statistically significant in any of the 4 reported specifications.
DISCUSSION The association between vending machine availability and BMI was statistically insignificant when students were grouped at aggregated levels by gender, race/ethnicity, or socioeconomic status. This is consistent with findings
Journal of Nutrition Education and Behavior Volume 47, Number 6, 2015
O’Hara and Haynes-Maslow 529
Table 2. Sweets, Salty Snacks, and Sweetened Beverage Availability and Purchases in Fifth and Eighth Grades
Food Type and Purchase
Fifth Grade n (%)
Eighth Grade n (%)
Sweets Can purchase sweets at school Times bought sweets at school, n/wk (mean SD)a Purchased sweets in school vending machinesb
1,138 (50.3) 1.65 3.55 26 (4.6)
1,533 (68.5) 1.46 3.16 98 (12.5)
Salty snacks Can purchase salty snacks at school Times bought salty snacks at school, n/wk (mean SD)a Purchased salty snacks in school vending machinesb
1,150 (46.4) 1.01 2.76 31 (8.6)
1,635 (73.8) 1.20 2.98 92 (14.0)
Sweetened beverages Can purchase soft drinks, fruit drinks, and sports drinks at school Times bought soft drinks, fruit drinks, and sports drinks at school, n/wk (mean SD)a Purchased soft drinks, fruit drinks, and sports drinks in school vending machinesb
882 (39.0) 0.94 2.58 131 (46.6)
1,423 (64.3) 1.09 2.57 300 (52.8)
a
Number of times food purchased per week at school is calculated for children who could purchase those foods at school; Purchased foods from school vending machines are calculated for students who purchased those foods at school.
b
by Van Hook and Altman.17 The predicted association between vending machine availability and higher BMI levels had greater statistical significance among Hispanic male and low-income children than other socioeconomic subgroups, although the finding among Hispanic subgroups
meals, likely owing to a shortage of time for purchasing and preparing food,20 language barriers that impede both children's and parents' dietary education,21 and taste preferences for nutrient-deficient foods.22 Hispanic male children may be more vulnerable to increases in BMI than are Hispanic
was insignificant with Bonferroni corrected critical values. Hispanic children from lowerincome families may have a greater association between vending machine availability and BMI than other subgroups because of challenges their parents confront preparing healthier
Table 3. First Difference Regressions for Change in Student’s BMI From Fifth to Eighth Grade Model 1 Hispanic Students (n ¼ 390) b Coefficient (SD) 2.46 (0.14)***
Model 2 Hispanic Male Students (n ¼ 210) b Coefficient (SD) 2.28 (0.20)***
Vending machine availability (categorical)
0.24 (0.20)
0.46 (0.24)*
0.40 (0.23)*
0.76 (0.33)**
Television watching, h/wk
0.03 (0.30)
0.05 (0.03)**
0.05 (0.02)**
0.09 (0.03)**
Independent Variables Constant
Exercises 20 min/d, d/wk Family receives SNAP (categorical) Eat breakfast with family, d/wk Urban (categorical) 2
0.05 (0.04) 0.78** (0.30) 0.02 (0.05) 1.25 (0.42)**
0.10 (0.06)* 0.48 (0.42) 0.10 (0.06)* 1.53 (0.54)**
Model 3 Low-Income Hispanic Students (n ¼ 264) b Coefficient (SD) 2.53 (0.17)***
Model 4 Low-Income Hispanic Male Students (n ¼ 134) b Coefficient (SD) 2.47 (0.30)***
0.04 (0.06)
0.10 (0.09)
0.70** (0.33)
0.49 (0.47)
0.03 (0.06)
0.10 (0.08)
1.35 (0.53)**
1.02 (0.88)
R
0.07
0.13
0.08
0.17
F test P
.01**
.03**
.01**
.02**
*P < .10; **P < .05; ***P < .002. Note: First difference estimates are derived from regression models that include changes in whether the child has a disability, lives with nonbiological parents, lives with a single mother, lives with a single father, days per week that the child eats dinner with the family, the ratio of income to federal poverty line, lives in a rural region, and the percent minority students enrolled in the child’s school as control variables. Parameter estimates for statistically insignificant variables are not reported.
530 O’Hara and Haynes-Maslow female children because, as reported in the ECLS-K dataset, they purchased salty snacks and sweetened beverages with a greater frequency at school. Other findings from this study reveal that television watching may be positively associated with BMI, which is consistent with evidence that sedentary lifestyles can contribute to increases in BMI.23 The positive parameter estimate associated with exercise in males could result from increased muscle mass. Consistent with other studies, living in an urban area may be associated with increased BMI, possibly owing to a lack of recreational venues and safe places to be active.24 Supplemental Nutrition Assistance Program enrollment and eating breakfast with the family had a negative association with BMI, possibly because these factors indicate greater food security and more stable eating patterns. The magnitude of the parameter estimates for the constant was similar across models 1–4 and ranged from 2.3 to 2.5.
IMPLICATIONS FOR RESEARCH AND PRACTICE Because of concern regarding the health impacts of children's diets, federal nutrition standards for competitive foods in schools, including those sold in vending machines, were updated on July 1, 2014. Implemented in the 2014–2015 academic year, nutrition standards for sweetened beverages—the most commonly purchased item in school vending machines—must not contain more than 60 calories per 12-oz serving. This is approximately 40% to 43% of the calories in a nondiet 12-oz soft drink. The health impacts of the new competitive food standards have yet to be evaluated. Although the calorie amounts for beverages are lower, such changes could result in increased sweetened beverage consumption if students' perceptions of serving sizes are altered (eg, fewer calories in a serving size may decrease the guilt of overconsuming calories from sweetened beverages).25 Thus, understanding whether changes in the school food environment have differential associations with chil-
Journal of Nutrition Education and Behavior Volume 47, Number 6, 2015 dren's health status by subgroup remains pertinent and highlights the value of research on this topic estimating associations at specific levels of subgroups.
10.
ACKNOWLEDGMENTS The work was undertaken when both authors were employed at the Union of Concerned Scientists.
REFERENCES 1. Ogden CL, Carroll MD, Kit BK, Flegal KM. Prevalence of childhood and adult obesity in the United States, 2011-2012. JAMA. 2014;311: 806-814. 2. Singh GK, Siahpush M, Kogan MD. Rising social inequalities of US childhood obesity, 2003-2007. Ann Epidemiol. 2010;20:40-52. 3. Skelton JA, Cook SR, Auinger P, Klein JD, Barlow SE. Prevalence and trends of severe obesity among US children and adolescents. Acad Pediatr. 2009; 9:322-329. 4. Sallis JF, Owen N, Fisher EB. Ecological models of heath behavior. In: Glanz K, Rimer BK, Viswanath K, eds. Health Behavior and Health Education: Theory, Research and Practice. 4th ed. San Francisco, CA: Josey-Bass; 2008. 5. Fernandes MM. The effect of soft drink availability in elementary schools on consumption. J Am Diet Assoc. 2008; 108:1445-1452. 6. Briefel RR, Crepinsek MK, Cabili C, Wilson A, Gleason PM. School food environments and practices affect dietary behaviors of U.S. public school children. J Am Diet Assoc. 2009;109(2 suppl):S91-S107. 7. Park S, Sappenfield WM, Huang Y, Sherry B, Bensyl DM. The impact of the availability of school vending machines on eating behavior during lunch: the youth physical activity and nutrition survey. J Am Diet Assoc. 2010; 110:1532-1536. 8. Rover AJ, Nansel TR, Wang J, Lannotti RJ. Foods sold in school vending machines are associated with overall student dietary intake. J Adolesc Health. 2011;48:13-19. 9. O’Toole TP, Anderson S, Miller C, Guthrie J. Nutrition services and foods and beverages available at school: re-
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
sults from the School Health Policies and Programs Study 2006. J Sch Health. 2007;77:500-521. Fox MK, Gordon A, Nogales R, Wilson A. Availability and consumption of competitive foods in US public schools. J Am Diet Assoc. 2009; 109(suppl 2):S57-S66. US Department of Agriculture and US Dept of Health and Human Services. Dietary Guidelines for Americans, 2010. 7th ed. Washington, DC: US Government Printing Office; 2010. Hu FB, Malik VS. Sugar-sweetened beverages and risk of obesity and type 2 diabetes: epidemiological evidence. Physiol Behav. 2010;100:47-54. Marshall TA, Levy SM, Broffitt B, et al. Dental caries and beverage consumption in young children. Pediatrics. 2003;112:e184-e191. Larson N, Story M. Are ‘‘competitive foods’’ sold at school making our children fat? Health Affairs. 2010;29: 430-435. Anderson PM, Butcher KM. Reading, writing, and refreshments: are school finances contributing to children’s obesity? J Human Resources. 2006;41: 467-494. Datar A, Nicosia N. Junk food in schools and childhood obesity. J Policy Analysis Management. 2012;31:312-337. Van Hook J, Altman CE. Competitive food sales in schools and childhood obesity: a longitudinal study. Sociol Educ. 2012;85:23-29. US Department of Health and Human Services. The 2007 HHS poverty guidelines. http://aspe.hhs.gov/poverty/07pov erty.shtml. Accessed November 15, 2014. US Department of Health and Human Services. The 2004 HHS poverty guidelines. http://aspe.hhs.gov/poverty/04po verty.shtml. Accessed November 3, 2014. Haynes-Maslow L, Parsons SE, Wheeler SB, Leone LA. Understanding barriers to fruit and vegetable consumption among low-income individuals: a qualitative study. Prev Chronic Dis. 2013;10:1202-1206. Gray VB, Cossman JS, Dodson WL, Byrd SH. Dietary acculturation of Hispanic immigrants in Mississippi. Salud Publica Mex. 2005;47:351-360. Batis C, Hernandez-Barrera L, Barquera S, Rivera JA, Popkin BM. Food acculturation drives dietary differences among Mexicans, Mexican
Journal of Nutrition Education and Behavior Volume 47, Number 6, 2015 Americans, and non-Hispanic whites. J Nutr. 2011;141:1898-1906. 23. Boone JE, Gordon-Larsen P, Adair LS, Popkin BM. Screen time and physical activity during adolescence: longitudinal
effects on obesity in young adulthood. Int J Behav Nutr Phys Act. 2007;4:26. 24. Wolch J, Jerrett M, Reynolds R, et al. Childhood obesity and proximity to urban parks and recreational resources:
O’Hara and Haynes-Maslow 531 a longitudinal cohort study. Health Place. 2011;17:207-214. 25. Wansink B, Chandon P. Can ‘‘low-fat’’ nutrition labels lead to obesity? J Market Res. 2006;43:605-607.
531.e1 O’Hara and Haynes-Maslow
CONFLICT OF INTEREST The authors have not stated any conflicts of interest.
Journal of Nutrition Education and Behavior Volume 47, Number 6, 2015